The BART study is comparing 4 resilience training regimens: paced breathing at 5 or 6 breaths per minute, each with or without HRV biofeedback. Participants are randomly assigned to 1 of these 4 regimens and asked to practice paced breathing at least 3 times a week for 6 weeks, and thereafter for 12 months (Table 1). Participants use a chest-belt HR sensor to acquire HRV measurements, including participants randomly allocated to no HRV biofeedback. While such participants cannot observe changes in HRV during training, continuous acquisition of HRV data in the background enables posttraining analyses of physiological responses to cognitive stress and paced breathing training.

The study begins with a setup day (day 0) on which individuals provide their consent to participate, install the app on their smartphone, complete baseline assessments, learn to use their HR monitor (Polar H7; Polar Electro, Bethpage, NY, USA), and practice 1 resilience training session, which includes a cognitive stress game. Special consideration is given to wearing the Polar HR sensor, linking the sensor to the BART app, and acquiring good-quality HR measurements. After this initial setup and instruction, participants execute all activities on their own for the duration of the study under the scheduling and direction of the BART app.

BART app design was governed by the study protocol and schedule of participant activities (Table 1), the prescribed resilience training and stressor regimens, the mechanisms for helping participants to complete the study activities, and incentives to encourage adherence. A suite of self-report measures (eg, anxiety, posttraumatic stress, sleep quality, resilience) are taken at baseline, with a subset taken quarterly and at 12 months. Scheduling of activities is provided via the app, along with incentives to support adherence over the initial 6 weeks and throughout the 12 months. Owing to the geographic distribution of study recruitment sites, participants enter the study incrementally, thereby allowing a small study team to recruit, take consent, and provide initial training at various locations over an extended period. Consequently, each participant’s protocol schedule is based on their personal study entry date.

Our previous work in predeployment stress inoculation training, HRV biofeedback, and mobile technologies for mindfulness-based stress reduction strongly influenced our design of the BART app [18,25,26]. Each of these studies involved stress relaxation training, a cognitive stressor, and HRV assessments. We reviewed our lessons learned from these studies to refine processes and incorporate new sensors and mobile technology in the BART app. Smartphone-delivered health and HRV biofeedback analyses of the prior Personal Health Intervention Toolkit for Duty research app [27] constituted the foundation for app development.

We implemented the BART app using the Personal Health Informatics and Intervention Toolkit (PHIT), a development framework geared to research-oriented mobile apps [28-30]. The PHIT framework eases app development for acquiring data, including self-report instruments, ecological momentary assessment diaries, cognitive tests, and game-like activities. For sensor data collection, PHIT supports intrinsic (eg, global positioning system, motion) and Bluetooth 4.0 data streams (eg, HR monitors). All data are tagged with study protocol, participant, date and time stamps, and other contextual information, then encrypted and stored locally in the app space. A virtual advisor provides a logic layer where analysis and planning take place. An activity manager schedules self-report and sensor data collection, intervention and training, alerts, incentive feedback, and behavior change according to the study protocol.

PHIT modules are implemented using XML and employ PHITScript to construct program logic and activate special app functions, such as collecting sensor data or scheduling notifications. Apps using the PHIT framework run locally without the need for an active internet connection. PHIT is based on Apache Flex (Apache Software Foundation) and AIR (Adobe Systems) technologies, which are both open source and widely used for mobile game development. A requirement for the BART study was that participants would use their own smartphones or tablets, necessitating app compatibility with both Android and iOS devices as provided by Adobe AIR.

All acquired and derived data are stored on the device in an encrypted SQLite (SQLite Consortium) database and periodically uploaded by the participant to a secure central data server. To eliminate financial burden, neither continuous internet access nor use of the participant’s cellular data plan is required. Rather, data are uploaded whenever Wi-Fi internet access is available, and at the participant’s direction and convenience, either via Wi-Fi or the participant’s cellular data plan.

The BART operational schema (Figure 1) centers on a participant activities menu with various tasks such as health assessments, resilience training, and data uploads. The activities menu (Figure 1 and Figure 2a) is updated daily by a schedule manager according to protocol specifications and personal progress, using labeling and icon references as defined by PHITScript programming. Each day, activities are listed or removed, and a local notification is posted to the participant as a reminder to complete their activities. The menu may also be updated by changing the icon to note an incomplete activity, removing a completed activity from the menu, or tagging a listed activity with key user information—for example, advising on the number of resilience trainings remaining to meet incentive payment requirements for the current study week.

Primary outcome measures acquired via the app are resilience (Connor-Davidson Resilience Scale), coping measures (Brief Coping Scale, Perceived Stress Scale, and Posttraumatic Growth Inventory), and sleep problems (Sleep Disturbance Scale). Secondary outcomes are mental health (measured by the PTSD Checklist, 7-item Generalized Anxiety Disorder scale, and Center for Epidemiologic Studies Depression Scale), physical health (Short Form Health Survey), and alcohol use problems (Alcohol Use Disorders Identification Test). Covariates are combat and deployment, recent tobacco and caffeinated beverage use, age, education, use of other relaxation techniques, and interest in learning relaxation skills. These measures, along with demographic information, were aggregated into a set of brief survey instruments to be completed at baseline (surveys 1-3), weekly, quarterly, and at 12 months.

Health assessments are administered via brief self-report instruments, typically with a single question per screen (Figure 2b and c). As the user advances through each assessment, a graphic indicator informs progress toward completion. At completion of each self-report measure or resilience practice exercise, an incentives manager records the earned incentive to the database (Figure 1). The user is then advised to upload data or defer the upload to a more convenient time. The activities menu may also be updated. For activities with HR sensor data streams, a Bluetooth interface manager links the sensor and receives beat-by-beat HR information for HRV analysis. Once initiated, this process executes autonomously in the background while the participant performs resilience training. The raw HR and derived HRV measures are provided for feedback display and saved in the app database.

The BART project employs real-time HRV analysis to provide physiological biofeedback during resilience training. Beat-by-beat heart intervals, also called interbeat intervals, are acquired continuously during each training session from a Bluetooth Low Energy (Bluetooth Special Interest Group) HR monitor. The raw interbeat intervals are streamed in real time to an HRV analysis module and stored in the app database to allow for subsequent offline quality review and analysis.

Three variants of HRV measures are determined using the Porges-Bohrer HRV analysis methodology [31,32]: respiratory sinus arrhythmia (RSA), low-frequency HRV, and wideband HRV. RSA reflects parasympathetic vagal activity for expected spontaneous breathing rates, whereas low-frequency HRV is thought to reflect sympathetic activity, as well as other cardiovascular regulatory systems. The wideband measure ensures that very low breathing rates during paced breathing exercises are properly measured. Multimedia Appendix 1 provides HRV data processing details.

We asked participants to practice resilience training 3 times each week for 6 weeks, a 2-step process comprising a 3-minute resting segment and a 5-minute resilience training segment. Each participant was randomly allocated after consent to receive 1 of 4 resilience training regimens: paced breathing at 5 or 6 breaths per minute, with or without HRV biofeedback. Before training, the participant is asked to be in a quiet place and put on the Polar H7 HR monitor. When the participant is ready, the HR monitor is activated and a beat-by-beat HR trend is displayed to check signal quality (Figure 3a). The participant reviews the HR trend and decides whether to proceed to resilience training or take measures, such as adjusting or moistening the chest strap sensor, to improve data quality. Resilience training begins with a 3-minute resting segment to relax the participant and establish baseline HRV measures. During this time, the participants may close their eyes or lightly focus on a series of peaceful landscapes that fade from one to another at 30 second intervals (Figure 3b). A narrator announces when each minute arrives to help reduce anxiety owing to waiting for the resting segment to finish.

For participants receiving resilience training without biofeedback, an animated ball is displayed as rising and falling upon a triangular graphic for paced breathing resilience training (Figure 3c top). Participants inhale as the ball rises and exhale while the ball falls, with ball movement set at 5 or 6 breaths per minute, with an inspiration to expiration ratio of 0.435 and an end-inspiration and end-expiration pause of 1.5 seconds. An audible tone with rising and falling pitch is played in synchrony with the rising and falling ball to allow for paced breathing with eyes closed.

For participants receiving resilience training with biofeedback, the animated ball and audible tones are rendered in similar fashion to that without biofeedback. Two modes of graphic biofeedback are provided: a trending HRV chart and a real-time dynamic HRV meter. The chart and meter are updated every 2 seconds against a color-coded background to show movement between calm (green) and stressful (red) psychophysiological states, reflecting higher and lower parasympathetic activation.

On study days 0 and 1, and after 6 weeks of training, participants complete an enhanced training regimen called the Training Game. The Training Game builds on the basic resilience training exercise by incorporating the Eriksen flanker task [33], a game-like cognitive stress exercise designed to elicit psychophysiological stress. The Eriksen flanker task heightens psychological stress by requiring attention, providing anticipation, and imposing conflict in higher brain function. As before, HRV is measured throughout the Training Game, thereby allowing for objective assessment of resilience before and after 6 weeks of training.

The Training Game begins with a 3-minute rest, followed by instructions on performing the Eriksen flanker task (Figure 4a). When ready, the participant begins the Eriksen flanker task, which presents a series of stimulus screens comprising a field of arrows pointing to the left or right, with a central arrow that may be congruent or incongruent in direction with the 4 bounding arrows on either side (Figure 4b). The 4 bounding arrows are randomly rendered as pointing left or right, resulting in 4 available stimulus combinations. At a random interval ranging from 1 to 3 seconds, 1 of the 4 combinations of the central and flanking arrows is selected at random and presented for 400 milliseconds. Participants have 2.7 seconds to respond by tapping the left or right button to indicate the direction of the central arrow. The Eriksen flanker task continues presenting stimuli for 4 minutes, then the BART app advances to a 3-minute poststress recovery phase of sitting at rest (Figure 4c). After recovery, resilience training is provided as previously described with paced breathing at 5 or 6 breaths per minute, with or without HRV biofeedback.

Adherence to study interventions and data collection is a common challenge in any research study, and especially when participants are asked to essentially perform the study on their own, albeit with app support. To support the frequency of resilience training (3 time per week), weekly assessments, and the yearlong period of follow-up assessments, the BART study design incorporated an incremental incentives approach to encourage participants to do scheduled activities and stay engaged across the duration. We were concerned, however, that manual monitoring of several hundred participants with time-shifted protocol schedules could be error prone and cause missed payments or awarding unearned payments. To mitigate such risk, we implemented incentive management to standardize both incentive qualification and automated distribution of incentive awards (Figure 5).

In concert with the BART protocol (Table 1), incentives are earned providing that the participant completes the requisite activities (Figure 5). Incentive criteria, along with labeling and monetary values, are coded in PHITScript and checked both daily and after the participant exits each activity module. Whenever the participant completes the criteria for a specific incentive, an incentive fulfillment request is stored to the local app database, and the in-app incentives table is updated to inform the user that an incentive has been earned and is pending award. When study data are transferred to the secure central data server, the incentive fulfillment request is noted for processing. Each night a procedure scans all participants for pending fulfillment requests, identifies unpaid incentives, emails a gift card code in the appropriate amount to participants, and tags the incentive payment as fulfilled.

To further support the incentive process, monitor payments, and validate the accuracy of the incentive management process, a report of incentives earned and paid is produced weekly so that study staff may check each participant’s incentive record. This report not only aids in confirming payments, but also helps to resolve any problems that may have been experienced by participants and supports monitoring adherence across all participants.

Ensuring privacy and security of data and on-device analysis results is an absolute necessity for ethical reasons, to meet human studies requirements, and to support data quality in the conduct of self-managed mobile research protocols. Participants are provided a randomized participant identification (ID), which uniquely links all acquired data to that individual without any personally identifiable information. They also enter a self-defined secret 4-digit personal identification number (PIN) to prevent access by other individuals. When in use, the app screen deactivates after a set period (eg, 2 minutes) of no interaction, and current data and activity is hidden. The 4-digit PIN must then be entered to unlock the screen and allow the participant to continue.

Implementing a study using an app installed on the participants’ devices requires app installation from a public app store. Since anyone might download and install the app, and possibly upload false data, we addressed this quality and security risk by including an app lockout requiring an unlock password. After participants consent to take part in the study, the unlock is revealed, the app installed, and the password entered to activate. This prevents extraneous persons from entering and corrupting the study.

All data are stored locally on the device in an encrypted SQLite database within the BART app, thereby permitting use without requiring a continuous internet connection. Data are stored using a 128-bit Advanced Encryption Standard algorithm with no personal identifying information. Data are periodically uploaded to a central secure data server whenever Wi-Fi internet access is available, thereby reducing use of the participants’ cellular data plans. Data are transferred using the secure https protocol and stored in a secure SQL server database, which is accessible only to authorized persons via user ID and password authentication.

To showcase how the BART app is being used and to present examples of the HRV measures during resilience training, we conducted a limited, interim review of user interactions and study data. Consequently, data presented here do not address the study hypotheses on the effectiveness of various training modes on building resilience. Analysis of training effectiveness on resilience and other outcome measurements will be addressed separately after completion of data collection.

We based data regarding app usage on participant rostering records and earned incentives reported. For each study activity (Table 1), we tallied a completed measure—that is, the number of participants who completed the activity and earned the corresponding incentive. Since each participant has a unique study calendar based on individual starting date, the activity schedule differs across participants. We therefore tallied the number of participants who were scheduled to perform each activity adjusted according to their individual start date (ranging from June 2017 to September 2018) until this analysis on October 1, 2018. Finally, we determined the ratio of completed to scheduled activities as a compliance measure for each required study activity. We calculated these analyses using Excel 2019 (Microsoft Corporation).

We reviewed the psychophysiological stress response during the cognitive stress and biofeedback training using the wideband biofeedback HRV measure across all segments (rest, stressor, recovery, and training). Our analysis was restricted to data taken during the first week of participation, before substantive resilience practice would yield any training effect.

We computed descriptive statistics for the subpopulation extracted for the HRV measurement examples, with categorical variables reported by frequencies and numeric variables by mean (SE). We analyzed grouped HRV data using a univariate general linear model and present the data graphically as mean (2 SE). We used unpaired t tests to evaluate changes in HRV for sequential training segments. We conducted statistical analyses using IBM SPSS Statistics 25 (IBM Corporation).